A method and system for optimizing operating conditions of a thermal power unit

Abstract

The present invention relates to the technical field of operating optimization of thermal power units, and discloses a method and system for optimizing operating conditions of thermal power units. The method includes obtaining first historical operating data and second historical operating data of the unit, performing anomaly processing on the data to obtain first processed data and second processed data; performing steady-state screening on the first processed data to obtain first steady-state data; using the fuzzy C-means clustering algorithm to perform operating condition division on the first steady-state data to obtain first operating condition categories, screening out the first optimal operating condition category from them, and extracting the corresponding operating parameters to form a first optimal operating condition table; performing online operating condition judgment and optimization on the second processed data, and updating the optimal operating condition table; the system includes a data acquisition module, a data processing module, a steady-state screening module, an operating condition division module, an optimal operating condition screening module, and an online operating condition optimization module; the present invention can effectively improve the operating efficiency and economy of thermal power units.

Description

Technical Field

[0001] The present invention relates to the technical field of operation optimization of thermal power units, and more specifically, to a method and system for optimizing operating conditions of thermal power units. Background Art

[0002] As the main power generation equipment in China's power system, the operating efficiency and economy of thermal power units directly affect the safe, stable and economic operation of the power system. How to optimize the operating conditions of thermal power units according to the actual operating status of the units and improve the unit efficiency is an urgent problem to be solved in thermal power plants.

[0003] Currently, there are some related technologies for optimizing the operation of thermal power units. For example, Chinese Patent with publication number CN109685264A discloses a method, device and computer equipment for optimizing the operation of thermal power units. This method establishes multiple subsystem optimization models based on operation historical data and the coupling degree and importance of the production process, and uses an optimization algorithm to iteratively optimize the models to quickly and effectively obtain the best process parameters for the operation of thermal power units. Another Chinese Patent with authorization announcement number CN108037748B discloses a method, device and equipment for optimizing the operation of thermal power units under all operating conditions. This method screens historical data through data mining to obtain economic operating conditions and economic rules that meet preset conditions, and supplements the missing content in the benchmark operating condition library through unit tests. Finally, the optimized economic rules are added to the benchmark operating condition library to guide the subsequent operation of the units.

[0004] Although the above patents provide some methods for optimizing the operation of thermal power units, they lack the ability of online real-time optimization and cannot perform dynamic optimization according to the real-time operating conditions of the units. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art, the present invention provides a method and system for optimizing operating conditions of thermal power units.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] A method for optimizing operating conditions of a thermal power unit, comprising:

[0008] Obtaining first historical operation data and second historical operation data of the unit, and performing anomaly processing on the obtained first historical operation data and second historical operation data to obtain first processed data and second processed data;

[0009] Perform steady-state screening on the first processed data to obtain the first steady-state data; use the fuzzy C-means clustering algorithm to divide the working conditions of the first steady-state data to obtain the first working condition categories, select the first optimal working condition category from the first working condition categories, and extract the corresponding operating parameters under the first optimal working condition category to form the first optimal working condition table;

[0010] Perform online working condition judgment and optimization on the second processed data to update the first optimal working condition table.

[0011] Further, the first historical operation data is the historical operation data of the unit in the most recent 12 months obtained from the real-time database, with a sampling interval of 1 minute; the second historical operation data is the operation data within the previous 24 hours obtained from the real-time database, with a sampling interval of 1 second.

[0012] Further, the abnormal processing of the obtained first historical operation data and second historical operation data includes:

[0013] Perform missing value detection item by item on the first historical operation data and the second historical operation data. If the data is missing, mark it as missing data. If the data is not missing, mark it as non-missing data; for the missing data, replace it with the average value of the 5 adjacent data before and after the adjacent moment; if there are 2 or more consecutive missing data in the adjacent data, mark it as unavailable data;

[0014] Perform physical threshold detection item by item on the non-missing data. Set the physical threshold range. If it exceeds the physical threshold range, mark it as the first abnormal data;

[0015] Perform statistical threshold detection item by item on the data that does not exceed the physical threshold range. Use 3 times the standard deviation as the statistical threshold. If the deviation from the data average value exceeds 3 times the standard deviation, mark it as the second abnormal data;

[0016] Perform noise detection item by item on the data that has not been detected as abnormal. Calculate the peak-to-peak value of the data. If the peak-to-peak value exceeds the set peak-to-peak threshold and the short-term variance of the data is higher than 3 times the long-term variance, mark it as the third abnormal data;

[0017] Eliminate the unavailable data, the first abnormal data, the second abnormal data, and the third abnormal data to obtain the processed first processed data and second processed data.

[0018] Further, the steady-state screening of the first processed data to obtain the first steady-state data includes:

[0019] Extract the time series data of the steady-state parameters from the first processed data. Divide the time series data of each steady-state parameter into sliding time windows with a length of L. The windows slide continuously at a step size of 1 sampling point;

[0020] The steady-state index SI is calculated for the data in each time window of each steady-state parameter;

[0021] Calculate SI for multiple steady-state parameters in each time window and take the maximum SI of each SI max As a comprehensive homeostatic index for this time window;

[0022] The average of the timestamps of all sampling points in the sliding time window is used as the comprehensive steady-state index SI max At the corresponding time t, the time series steady-state exponential curve SI is obtained. max (t);

[0023] Set the first steady-state threshold SI th , find out SI max All the values in (t) less than SI th Time interval T s , and T s If the interval is greater than the interval threshold T, the interval is determined to be a steady-state interval; and the first processed data in the steady-state interval is extracted as the first steady-state data.

[0024] Furthermore, the first steady-state data is divided into the first operating condition category by the operating condition classification, which includes:

[0025] Performing data standardization processing on the first steady-state data to obtain standardized steady-state data;

[0026] Initialize the parameters of the fuzzy C-means clustering algorithm, including the number of clusters C, the membership matrix U, the cluster center V, and the maximum number of iterations MAX ITER , objective function termination threshold ε;

[0027] Based on the standardized steady-state data, the membership matrix U and cluster center V are iteratively optimized;

[0028] According to the final membership matrix U, each data point is divided into the category with the largest membership, forming C working condition clusters, namely the first working condition category;

[0029] The iterative optimization of the membership matrix U and the cluster center V includes:

[0030] Based on the current membership matrix U, update the cluster centers V of each category k , k is the index of the number of cluster categories, V k represents the cluster center of the kth class;

[0031] Based on the current cluster center V, update the membership U of the data point to each category kj , U kj Indicates the membership of the jth sample to the kth class;

[0032] Calculate the objective function J and determine whether J is less than ε or the number of iterations reaches MAX ITER If so, generate the final membership matrix U; if not, continue the iteration.

[0033] Further, the screening of the first optimal operating condition category from the first operating condition category includes:

[0034] For each operating condition category, calculate its various performance indicators;

[0035] Perform a weighted sum of the various performance indicators to obtain the comprehensive performance evaluation value of each operating condition category;

[0036] Select the operating condition category with the optimal comprehensive performance evaluation value as the first optimal operating condition category.

[0037] Further, the online operating condition judgment and optimization for the second processed data include:

[0038] Perform a steady-state judgment on the second processed data. If it is in a steady state, perform an operating condition division on the second processed data to obtain the second operating condition category; compare the second operating condition category with the first optimal operating condition table. If the second operating condition category is better than the first optimal operating condition category, update the first optimal operating condition table with the operating parameters corresponding to the second operating condition category; if the second operating condition category is not better than the first optimal operating condition category, keep the first optimal operating condition table unchanged. After waiting for 24 hours, store the latest 24-hour data as the new second historical operating data in the real-time database and perform the next round of online operating condition judgment and optimization;

[0039] If it is in a non-steady state, keep the first optimal operating condition table unchanged. After waiting for 24 hours, store the latest 24-hour data as the new second historical operating data in the real-time database and perform the next round of online operating condition judgment and optimization.

[0040] A system for optimizing the operating conditions of a thermal power unit, which is used to implement the above-mentioned method for optimizing the operating conditions of a thermal power unit. The system includes:

[0041] Data acquisition module: used to obtain the first historical operating data and the second historical operating data of the unit;

[0042] Data processing module: used to perform anomaly processing on the obtained first historical operating data and second historical operating data to obtain the first processed data and the second processed data;

[0043] Steady-state screening module: used to perform steady-state screening on the first processed data to obtain the first steady-state data;

[0044] Operating condition division module: used to perform operating condition division on the first steady-state data by using the fuzzy C-means clustering algorithm to obtain the first operating condition category;

[0045] Optimal operating condition screening module: used to screen out the first optimal operating condition category from the first operating condition category, and extract the corresponding operating parameters under the first optimal operating condition category to form the first optimal operating condition table;

[0046] Online operating condition optimization module: used to perform online operating condition judgment and optimization on the second processed data, and update the first optimal operating condition table.

[0047] An electronic device includes a memory, a central processing unit, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, the steps in the above-mentioned method for optimizing the operating conditions of a thermal power unit are implemented.

[0048] A computer-readable storage medium stores a computer program, and when the computer program is executed, the steps in the above-mentioned method for optimizing the operating conditions of a thermal power unit are implemented.

[0049] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0050] By acquiring and processing the historical operation data of the unit, the present invention can timely discover and eliminate abnormal data, ensuring the accuracy and reliability of the data; using the fuzzy C-means clustering algorithm to divide the steady-state data into operating conditions can accurately identify the operating states of the thermal power unit under different operating conditions, and screen out the optimal operating conditions, thereby improving the operating efficiency of the unit.

[0051] By performing abnormal processing on the first and second historical operation data, the present invention can effectively reduce data noise and errors, avoid unnecessary maintenance and repairs caused by data problems, thereby reducing the operating cost; through online operating condition judgment and optimization, the optimal operating condition table can be updated in real time, ensuring that the thermal power unit is always in the best operating state, reducing energy consumption and equipment wear.

[0052] The present invention adopts advanced data processing methods, including missing value detection, physical threshold detection, statistical threshold detection, and noise detection, which can comprehensively and efficiently process data, improving the accuracy and efficiency of data processing; using the sliding time window and steady-state index calculation method can accurately screen out the steady-state data, ensuring the accuracy of subsequent operating condition division.

[0053] Through the fuzzy C-means clustering algorithm, the present invention can adaptively adjust the clustering center and membership matrix under different operating conditions, ensuring the accuracy and robustness of the operating condition division; the online operating condition optimization module can perform judgment and optimization on new data in real time, with strong adaptability, and can cope with various changes in the operation of the thermal power unit.

[0054] By weighted summation of the performance indicators for each working condition category, the present invention can accurately evaluate the comprehensive performance of each working condition, ensuring that the selected optimal working condition has the best operating performance; this method can effectively apply the performance evaluation results to actual operation, guide the optimization scheduling and operation management of thermal power units, and improve the overall operating effect.

[0055] The system in the present invention includes multiple modules such as data acquisition, data processing, steady-state screening, working condition division, optimal working condition screening, and online working condition optimization. Each module cooperates with each other to form a complete working condition optimization system; the modular design of the system makes it have good scalability and flexibility, and can be adjusted and optimized according to actual needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0057] Figure 1 It is a principle flow chart of a method for optimizing working conditions of a thermal power unit operation in the present invention;

[0058] Figure 2 It is a principle flow chart of a method for processing abnormal data of the first historical operation data to obtain the first processed data in a method for optimizing working conditions of a thermal power unit operation in the present invention;

[0059] Figure 3 It is a method flow chart for dividing working conditions of the first steady-state data in a method for optimizing working conditions of a thermal power unit operation in the present invention;

[0060] Figure 4 It is a method flow chart for iteratively optimizing the membership matrix U and the clustering center V in a method for optimizing working conditions of a thermal power unit operation in the present invention;

[0061] Figure 5 It is a method flow chart for forming the first optimal working condition table from the first working condition category in a method for optimizing working conditions of a thermal power unit operation in the present invention;

[0062] Figure 6 It is a functional module diagram of a system for optimizing working conditions of a thermal power unit operation in the present invention;

[0063] Figure 7 It is a schematic structural diagram of an electronic device provided by an embodiment of the present application;

[0064] Figure 8Schematic diagram of the structure of a computer-readable storage medium provided by an embodiment of the present application. Detailed implementation manners

[0065] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0066] Embodiment 1

[0067] Please refer to Figure 1 As shown, this embodiment provides a method for optimizing the operating conditions of a thermal power unit, including:

[0068] Step S1000, obtaining the first historical operation data and the second historical operation data of the unit, and performing anomaly processing on the obtained first historical operation data and the second historical operation data to obtain the first processed data and the second processed data;

[0069] Further, step S1000 includes:

[0070] Step S1100, obtaining the historical operation data of the unit in the most recent 12 months from the real-time database as the first historical operation data, with a sampling interval of 1 minute; obtaining the operation data within the previous 24 hours from the real-time database as the second historical operation data, with a sampling interval of 1 second;

[0071] Specifically, obtaining the historical operation data of the unit in the most recent 12 months from the real-time database through the data interface API. The data items include main steam pressure, main steam temperature, reheated steam temperature, generator power, furnace negative pressure, static pressure at the inlet of the induced draft fan, flue gas temperature at the outlet of the economizer, exhaust gas temperature, flame brightness, unit heat consumption, unit coal consumption, unit power generation coal consumption, boiler efficiency, auxiliary power, lower calorific value of the coal entering the furnace, sulfur content of the coal entering the furnace, volatile matter of the coal entering the furnace, nitrogen oxide emission concentration, sulfur dioxide emission concentration, etc. The sampling interval is 1 minute and it is stored as the first historical operation data; obtaining the operation data of the unit within the previous 24 hours from the real-time database through the same data interface API. The data items are the same as those of the first historical operation data, with a sampling interval of 1 second and it is stored as the second historical operation data. Using the historical data of 12 months is to cover various operating conditions of the unit under different loads, different coal types, and different environmental conditions, making the condition division more comprehensive; while using the operation data within 24 hours is to obtain the current condition information of the unit and provide a basis for condition optimization.

[0072] A real-time database is a database system specifically designed for real-time data processing and analysis, capable of capturing, storing, processing, and analyzing data simultaneously as it is generated. The establishment of a real-time database generally involves selecting an appropriate software platform, configuring hardware resources, defining a data model, and establishing data interfaces. Specifically, a real-time database automatically collects data from various data sources, such as sensors and control systems, through a data interface API, enabling low-latency data processing and high-throughput storage. Such a database system can provide real-time or near-real-time data access and analysis, ensuring the timeliness of data. In the method for optimizing the operating conditions of a thermal power unit, through the real-time database, historical operating data of the unit for the most recent 12 months and operating data within the previous 24 hours can be obtained, and are stored and processed at sampling intervals of 1 minute and 1 second respectively. These data include key parameters such as main steam pressure, temperature, and generator power, covering the full range of operating conditions of the unit and capable of reflecting the operating conditions under different loads, different coal types, and different environmental conditions. By using the real-time database, automatic collection and fusion of the unit's historical operating data and real-time operating data can be achieved, providing a solid data foundation for condition analysis and optimization, and ensuring the comprehensiveness and timeliness of data. Therefore, the real-time database plays a key role in this application scenario, providing a unified data access interface, shielding the heterogeneity of the underlying systems, and improving the efficiency of data collection and processing.

[0073] The automatic collection of the unit's historical operating data and real-time operating data is achieved by using the data interface API. This API accesses the real-time databases of each control system in the power plant in a unified manner, shielding the heterogeneity of the underlying systems and facilitating data fusion. For historical data, a time span of 12 months is selected, which can cover the full range of operating conditions of the unit, including operating data under different loads, different coal types, and different environmental conditions, thereby improving the comprehensiveness of condition division. For real-time data, a 24-hour time window is selected. On the one hand, sufficient data samples can be obtained, and on the other hand, computational efficiency is also taken into account to avoid excessive data volume affecting timeliness. The sampling frequencies of the two types of data are also set differently according to actual needs: historical data uses minute-level sampling, which can meet the accuracy requirements of statistical analysis; real-time data uses second-level sampling, which can achieve dynamic tracking of the unit's operating conditions. In addition, the selection of data items covers key parameters reflecting the unit's operating conditions, including thermal parameters, electrical parameters, fuel parameters, environmental protection parameters, etc. The fusion of multiple parameters can improve the distinguishability of condition characteristics. Step S1100 efficiently collects multi-source heterogeneous data of the unit's full operating conditions and performs appropriate spatio-temporal granularity settings, laying a solid data foundation for subsequent condition analysis.

[0074] Step S1200, perform abnormal data processing on the first historical operating data and the second historical operating data respectively to obtain the first processed data and the second processed data;

[0075] Further, as Figure 2 shown, step S1200 includes:

[0076] Step S1210: Detect missing values item by item for the first historical operation data and the second historical operation data. If the data is missing, mark it as missing data; if the data is not missing, mark it as non-missing data. For the missing data, replace it with the average value of the 5 adjacent data at the adjacent moments before and after. If there are 2 or more consecutive missing data in the adjacent data, mark it as unavailable data;

[0077] Step S1220: Detect physical thresholds item by item for the non-missing data, set the physical threshold range of this measuring point. If it exceeds the physical threshold range, mark it as the first abnormal data;

[0078] Step S1230: Detect statistical thresholds item by item for the data that does not exceed the physical threshold range in step S1220. Take 3 times the standard deviation as the statistical threshold. If the deviation from the data average value exceeds 3 times the standard deviation, mark it as the second abnormal data;

[0079] Step S1240: Detect noise item by item for the data that has not detected abnormalities in step S1230, calculate the peak-to-peak value of the data. If the peak-to-peak value exceeds the set peak-to-peak threshold and the short-term variance of the data is higher than 3 times the long-term variance, mark it as the third abnormal data;

[0080] Step S1250: Eliminate the unavailable data, the first abnormal data, the second abnormal data, and the third abnormal data detected in steps S1210 to S1240 to obtain the processed first processed data and the second processed data.

[0081] Specifically, step S1200 adopts a multi-abnormality detection method to achieve the layer-by-layer identification and elimination of missing values, over-limit values, outlier values, and noise values. First, perform missing value detection to ensure the integrity of the data. Replace sporadic missing values with the mean value of adjacent moment data, which can repair the data to a certain extent, while directly mark continuously large amounts of missing data as unavailable to prevent the introduction of errors.

[0082] Exemplarily, assume there is a set of time series data, and some data points are missing:

[0083]

[0084] In this example, the data points of t3, t4, and t5 are missing, and they are three consecutive missing points; the data point of t8 is missing, but its adjacent data points (t7 and t9) exist. Therefore, it can be replaced by the average value of the five data points before and after the adjacent time points. For t3, t4, and t5, since their adjacent data points (t2 and t6) cannot provide enough valid data to calculate the average value (because the adjacent t3, t4, and t5 are continuously missing themselves), these data points are marked as unavailable data.

[0085] Then, physical threshold detection is performed. Hard upper and lower limits of parameters are set according to the physical characteristics of the unit. Data outside the physical threshold range is significantly unreasonable and needs to be excluded. The setting of the physical threshold range needs to be based on the design parameters of the unit, such as the rated steam parameters of the boiler and the rated power of the steam turbine. Then, the 3σ criterion is used for statistical threshold detection to identify outliers that deviate too much from the average value. The 3σ criterion assumes that the data follows a normal distribution, and the probability of data falling outside 3 times the standard deviation is very small and can be regarded as abnormal. Finally, noise detection is carried out, and the violent volatility of the data is judged through the peak-to-peak value and short-term variance. Although the amplitude of the noise data may not be large, it changes quickly and needs to be identified from the perspective of signal stability. Through step-by-step screening, finally clean processed data is obtained. The first processed data represents the historical data of each working condition of the unit, and the second processed data represents the real-time data of the current working condition. Abnormal detection avoids the impact of dirty data on modeling analysis and is a key link in data preprocessing.

[0086] Step S1200 adopts a multi-level and multi-index abnormal data detection method, which can comprehensively identify various abnormal situations in the working condition data. Missing value detection repairs the integrity of the data and reduces information loss; physical threshold eliminates the wrong data that obviously violates the unit characteristics; statistical threshold identifies the accidental outliers that deviate from the normal level; noise detection removes high-frequency interference and equipment fault data. By comprehensively using mechanism knowledge and statistical means, finally a high-quality working condition data set is obtained, thus providing reliable data support for working condition optimization. At the same time, the abnormal detection method is separately applied to the first and second types of data, which not only ensures the data quality of offline working condition analysis but also takes into account the real-time requirements of online optimization. After the data cleaning in step S1200, the success rate and convergence speed of working condition optimization will be significantly improved.

[0087] In step S2000, the first processed data is subjected to steady-state screening to obtain the first steady-state data; the fuzzy C-means clustering algorithm is used to divide the first steady-state data into working condition categories to obtain the first working condition category, the first optimal working condition category is screened out from the first working condition category, and the corresponding operating parameters under the first optimal working condition category are extracted to form the first optimal working condition table;

[0088] Furthermore, step S2000 includes:

[0089] Step S2100, perform steady-state screening on the first processed data to obtain the first steady-state data;

[0090] Further, step S2100 includes:

[0091] Step S2110, extract the time series data of the steady-state parameters from the first processed data, divide the time series data of each steady-state parameter into sliding time windows with a length of L, and the windows slide continuously with a step size of 1 sampling point, where L is 2 - 60 min;

[0092] Step S2120, calculate the steady-state index SI for the data within each time window of each steady-state parameter;

[0093] Step S2130, calculate SI for each of the multiple steady-state parameters within each time window respectively, and take the maximum value SI max as the comprehensive steady-state index of this time window;

[0094] Step S2140, take the average value of the timestamps of all sampling points within the sliding time window as the moment t corresponding to the comprehensive steady-state index SI max to obtain the time series steady-state index curve SI max (t);

[0095] Step S2150, set the first steady-state threshold SI th , and find all time intervals T max in SI th that are less than SI s , and if T s is greater than the interval threshold T, then determine that this interval is a steady-state interval; extract the first processed data within the steady-state interval as the first steady-state data.

[0096] Specifically, step S2100 adopts the methods of sliding time window and comprehensive steady-state index to realize the automatic identification of the multi-parameter steady-state conditions of the unit. Traditional steady-state determination often uses the single-parameter over-threshold method, which is difficult to describe complex working conditions, while this method integrates multiple key parameters reflecting the unit state and comprehensively examines the stability degree of the unit. First, identify the time periods reflecting the stable operation state of the unit from the massive first processed data, and extract the time series of each process parameter within these steady-state periods to form the time series data of each steady-state parameter; the steady-state parameters include main steam pressure, main steam temperature, reheater steam temperature, generator power, furnace negative pressure, inlet static pressure of induced draft fan, flue gas temperature at the outlet of economizer, exhaust gas temperature, flame brightness, etc.

[0097] Construct a time window for each steady-state parameter. The selection of the window length needs to balance the timeliness and reliability of the steady-state criterion. The value of L is determined according to the time scale of the unit operating condition change and can be selected from 2 to 60 minutes. If it is too short, it is vulnerable to transient interference; if it is too long, misjudgment may occur. Then, construct a dimensionless steady-state index SI using the mean and standard deviation of the data within the window. The smaller the SI value, the more stable the parameter is within the window. The definition of SI refers to the capability index in statistical process control, but the upper and lower specification limits are replaced by the standard deviation to make it applicable to steady-state determination. Extract the comprehensive steady-state index SI through the maximum value of the multi-parameter SI max , which can judge whether all parameters simultaneously meet the stable state and avoid the limitations of the single-parameter criterion. Use the sliding window technique to obtain the continuous curve of SI max , and then perform threshold segmentation on it to obtain the interval where the steady-state duration exceeds the given threshold. The historical data within the interval is the steady-state operating condition data and can be used for subsequent operating condition clustering. Step S2100 makes full use of the dynamic characteristics and correlations of the operating condition data, depicts the steady-state characteristics of the unit through quantitative indicators, is more objective and comprehensive than traditional methods, and provides a high-quality sample set for online operating condition optimization

[0098] The method for calculating the steady-state index SI includes:

[0099]

[0100] Among them:

[0101] x i : The value of the i-th sample data point, which is extracted from the time series data of the steady-state parameter;

[0102] μ: The average value of all sample data points within the time window, representing the average level of the data within the time window and reflecting the central tendency of the data;

[0103] σ: The standard deviation of all sample data points within the time window, indicating the degree of dispersion or volatility of the data and reflecting the stability of the data;

[0104] w i : The weight of the i-th sample data point. The weight can be assigned according to the importance of the sample data point to increase the flexibility of the formula. Usually, it can be set to 1. If some data points are more important than others, higher weights can be assigned to them;

[0105] κ: The adjustment coefficient, which is used to adjust the sensitivity of the steady-state index to volatility to make the formula more flexible in different applications;

[0106] ν: The change rate of the data points within the time window, and the calculation method is Reflects the dynamics and volatility of the data, helping to identify situations of short-term drastic changes;

[0107] N: The number of sample data points within the time window.

[0108] When the data point x i undergoes a drastic change, |x i - μ| will also increase accordingly, thus causing SI to increase, indicating a decrease in stability; the change in the average value μ will affect the calculation of |x i - μ|, thereby affecting the value of SI. If μ increases while x i remains unchanged, the overall SI may increase; an increase in the standard deviation σ means an increase in the volatility of the data, resulting in a decrease in SI, reflecting a deterioration in the stability of the data; increasing the weight w i of certain data points will increase the influence of these points on SI, making the formula more flexible; increasing the adjustment coefficient κ will increase the sensitivity of the formula to the change rate ν, enabling the stability index to better reflect the drastic changes in the data.

[0109] The weight w i and the adjustment coefficient κ, the numerical ranges and selections of these two parameters have an important impact on the calculation of the stability index. For different application scenarios and data characteristics, the preferred numerical ranges are as follows:

[0110] The weight w i has a preferred numerical range of 0 ≤ w i ≤ 2;

[0111] When w i = 0, it means ignoring this data point;

[0112] When w i = 1, it means the importance of this data point is the standard value;

[0113] When w i > 1, it means increasing the importance of this data point.

[0114] The importance of data points is usually obtained based on historical data analysis. For key data points (e.g., key parameters reflecting the unit operating conditions), their weights can be appropriately increased. Excessive weights may lead to an over - influence of certain data points on the stability index and need to be set with caution.

[0115] The preferred numerical range of the adjustment coefficient κ is 0 ≤ κ ≤ 1;

[0116] When κ = 0, it means ignoring the influence of the change rate on the stability index;

[0117] When κ = 1, it means maximizing the influence of the change rate on the stability index.

[0118] The adjustment coefficient is determined based on the actual requirements of the unit operation and historical data analysis. A larger κ indicates greater sensitivity to data fluctuations and is applicable to scenarios with high requirements for steady state; a smaller κ is applicable to scenarios with a high tolerance for fluctuations.

[0119] This formula comprehensively considers the degree of data dispersion (standard deviation), enabling the steady-state index to comprehensively reflect the stability of the data; by adjusting the coefficient κ and the weight w i , the sensitivity of the formula and the importance of data points can be adjusted according to actual needs; by calculating the average value and standard deviation of the data, the formula can effectively smooth out accidental fluctuations in the data and provide a more stable steady-state index.

[0120] Exemplarily, assume there is the following first processed data, which includes three steady-state parameters: main steam pressure, main steam temperature, and generator power; the data sampling interval is 1 minute, and the data length is 10 time points.

[0121] The original data is:

[0122]

[0123] Set the window length L = 3 minutes and slide with a step size of 1 sampling point. The windows are as follows:

[0124] Window 1: T1 - T3

[0125] Window 2: T2 - T4

[0126] Window 3: T3 - T5

[0127] Window 4: T4 - T6

[0128] Window 5: T5 - T7

[0129] Window 6: T6 - T8

[0130] Window 7: T7 - T9

[0131] Window 8: T8 - T10

[0132] According to Calculate the SI of each steady-state parameter within each time window, w i Take 1, and κ takes 0.5;

[0133] Window 1 (T1 - T3):

[0134] Main steam pressure data points: [9.5, 10.0, 9.8];

[0135] Average value (μ): 9.7667;

[0136] Standard deviation (σ): 0.2517;

[0137] Rate of change (ν): 0.35;

[0138] Steady-state index of main steam pressure:

[0139]

[0140] Main steam temperature data points: [540, 542, 541];

[0141] Mean value (μ): 541;

[0142] Standard deviation (σ): 1;

[0143] Rate of change (ν): 1.5;

[0144] Steady-state index of main steam temperature:

[0145]

[0146] Data points: [100, 102, 101];

[0147] Mean value (μ): 101;

[0148] Standard deviation (σ): 1;

[0149] Rate of change (ν): 1.5;

[0150] Steady-state index of generator power:

[0151]

[0152] … (calculate SI for each window);

[0153] Calculate SI for multiple steady-state parameters within each time window respectively, and take the maximum value SI of each SI max as the comprehensive steady-state index of this time window;

[0154] Window 1 (T1 - T3): SI max = max(1.807, 1.143, 1.143) = 1.807;

[0155] … (calculate in the same way for each window);

[0156] Plot the comprehensive steady-state index SI of each window max as a time-series steady-state index curve according to time points, and obtain SI max (t).

[0157] Set the first steady-state threshold SI th , for example, set it to 2.0, and find all values in SI max (t) that are less than SI thIf the time interval is such that the interval length is greater than the interval threshold T, then this interval is determined to be a steady state interval. Finally, first processed data is extracted from the steady state interval to obtain first steady state data.

[0158] Step S2200: Use the fuzzy C-means clustering algorithm to perform working condition classification on the first steady state data to obtain the first working condition category;

[0159] Furthermore, as Figure 3 shown, step S2200 includes:

[0160] Step S2210: Perform data standardization processing on the first steady state data to obtain standardized steady state data;

[0161] Step S2220: Initialize the parameters of the fuzzy C-means clustering algorithm, including the number of clusters C, the membership matrix U, the cluster centers V, the maximum number of iterations MAX ITER and the termination threshold ε of the objective function;

[0162] Step S2230: Based on the standardized steady state data, iteratively optimize the membership matrix U and the cluster centers V;

[0163] Furthermore, as Figure 4 shown, step S2230 includes:

[0164] Step S2231: Based on the current membership matrix U, update the cluster centers V of each category k , where k is the index of the number of clustering categories, and V k represents the cluster center of the k-th category;

[0165] Step S2232: Based on the current cluster centers V, update the membership degrees U of the data points to each category kj , where U kj represents the membership degree of the j-th sample belonging to the k-th category;

[0166] Step S2233: Calculate the objective function J, and determine whether J is less than ε or the number of iterations has reached MAX ITER . If so, generate the final membership matrix U and execute step S2240. If not, return to step S2231 to continue the iteration.

[0167] Step S2240: According to the final membership matrix U, divide each data point into the category with the largest membership degree to form C working condition clusters, that is, the first working condition category.

[0168] Specifically, in step S2200, the fuzzy C-means (FCM) clustering algorithm is used to partition the steady-state operating condition data. The FCM algorithm is an unsupervised learning method that adaptively discovers the cluster structure of the data by optimizing the objective function. Different from traditional hard-partitioning clustering methods (such as K-means), FCM is based on fuzzy set theory and allows data points to belong to multiple categories. The membership degree reflects the similarity between the data points and each cluster center. This soft-partitioning method can better describe the transition characteristics between complex operating conditions, making the operating condition partitioning more reasonable and natural.

[0169] Before applying the FCM algorithm, it is necessary to standardize the original data. The unit parameters have different physical dimensions and numerical ranges. Directly using the original data may cause some parameters to dominate the distance calculation and affect the clustering effect. Through data standardization, each parameter can be mapped to the same numerical interval (such as [0, 1]), eliminating the dimension difference and making the contributions of different parameters to clustering equivalent. Common standardization methods include maximum-minimum standardization, zero-mean standardization, decimal scaling standardization, etc.

[0170] The FCM algorithm needs to preset some parameters, including the expected number of operating condition clusters C, the membership matrix U, the cluster centers V, the iteration control parameters MAX ITER and ε, etc. Among them, C can be determined according to the actual operating experience of the unit and historical data analysis. Usually, 3 - 10 categories are selected; U and V can be randomly initialized or initialized with other clustering results (such as K-means); the parameter m is the fuzzy exponent, which controls the "softening" degree during the clustering process; MAX ITER and ε are used to control the convergence of the algorithm and prevent excessive invalid iterations. The selection of these parameters has a certain impact on the clustering results and needs to be repeatedly tested and optimized.

[0171] The core of FCM is to alternately optimize the membership matrix U and the cluster centers V. Each iteration consists of two steps: First, fix the membership matrix U and update the cluster centers based on the current class partition; then, fix the cluster centers V and update the membership matrix based on the current center positions. These two steps are alternated, making the cluster centers continuously move towards the data-dense regions. At the same time, the membership degrees of the data points are updated according to the distances to the cluster centers, and finally converge to a local optimal solution. The objective function J measures the sum of the weighted distances from the data points to the centers of their respective classes. The smaller J is, the better the clustering effect. When the change in the objective function is less than the threshold ε, or the number of iterations exceeds MAX ITER the algorithm terminates.

[0172] After the algorithm converges, for each data point, the category with the largest membership degree is selected as its working condition. In this way, the original steady-state working condition data set is divided into C non-overlapping subsets, and each subset represents a typical unit operation condition. This division makes full use of the intrinsic structure information of the data and can adaptively extract the common features of different working conditions. Compared with manual division, FCM clustering is more objective and comprehensive, and can discover potential working condition patterns in the data.

[0173] The mathematical formula of the objective function J includes:

[0174]

[0175] Where:

[0176] n: The number of samples of the standardized steady-state data;

[0177] C: The number of clustering categories;

[0178] U kj : The membership degree of the j-th sample belonging to the k-th category, initialized to a random value, and obtained through iterative optimization;

[0179] m: The fuzziness index of the membership matrix U, usually in the range of [1.5, 2.5];

[0180] Y j : The j-th sample data;

[0181] V k : The clustering center of the k-th category, initialized to a random value, and obtained through iterative optimization;

[0182] ∥·∥: The L2 norm (Euclidean distance) of the vector, used to measure the distance between the sample and the clustering center;

[0183] λ: The distance weighting coefficient, controlling the influence degree of the distance on the membership degree, usually in the range of [0.1, 10], adjusted according to the clustering effect;

[0184] δ: The scale parameter of the Gaussian kernel function, controlling the rate of distance attenuation, usually in the range of [0.1, 1], selected according to the data distribution characteristics;

[0185] η: The category balance penalty coefficient, controlling the penalty degree for category imbalance, usually in the range of [0.01, 0.1], adjusted according to the category distribution;

[0186] n k : The number of samples in the k-th category, obtained by summing the membership matrix U;

[0187] p: The power exponent of the class balance penalty, which controls the non-linearity of the penalty. It usually ranges from [0.5, 2] and is selected according to the severity of class imbalance.

[0188] This objective function consists of two parts: The first term is the objective function of traditional fuzzy C-means clustering, which is used to minimize the sum of weighted distances between samples and the cluster center with the highest membership degree; the second term is the newly introduced class balance penalty term, which is used to penalize the class imbalance situation and promote the balance of the number of samples in each class.

[0189] In the first term, a Gaussian kernel function is introduced to weight the distance, making samples closer to have a greater impact on the membership degree, while the impact of samples farther away is weakened, improving the robustness of clustering and the tolerance to noise. As the distance between the sample and the cluster center increases, the value of the Gaussian kernel function will decay rapidly, making the contribution of samples far from the cluster center to the objective function smaller.

[0190] In the second term, the class balance penalty term will increase as the class imbalance degree intensifies. When the number of samples in a certain class is much smaller than that in other classes, the value of the penalty term will become larger, forcing the optimization process to adjust in the direction of increasing the number of samples in this class to achieve class balance. The power exponent p in the penalty term controls the non-linearity of the penalty. The larger p is, the greater the penalty for class imbalance.

[0191] By minimizing this objective function, while achieving sample clustering, the balance of classes can be taken into account, and a more reasonable and meaningful clustering result can be obtained. In the iterative optimization process, the membership matrix U and the cluster centers V will be continuously updated until the objective function value converges or reaches the maximum number of iterations. Generally speaking, based on traditional fuzzy C-means clustering, this objective function introduces distance weighting and class balance penalty, improving the robustness and class balance of clustering, and helping to obtain more accurate and reliable results in the clustering analysis of the working condition data set.

[0192] Exemplarily, an example is given to illustrate the process of working condition division of the fuzzy C-means clustering algorithm.

[0193] Suppose there is the following first steady-state data, which contains 10 data samples, and each sample has 3 feature variables: main steam pressure, main steam temperature, and generator power.

[0194] Sample Number Main Steam Pressure (MPa) Main Steam Temperature (°C) Generator Power (MW) 1 17.1 535 300 2 16.8 530 290 3 17.5 540 305 4 17.3 538 308 5 16.5 525 285 6 17.0 532 295 7 16.9 529 293 8 17.4 542 310 9 16.7 526 288 10 17.2 537 303

[0195] The data after standardizing the original data is as follows:

[0196]

[0197]

[0198] Initialize the clustering parameters, set the number of clusters C = 3, the maximum number of iterations MAX ITER = 100, and the objective function threshold ε = 0.01; the membership matrix U and the cluster centers V are randomly initialized.

[0199] The initial membership matrix U is as follows:

[0200] Sample Number Category 1 Category 2 Category 3 1 0.3 0.5 0.2 2 0.7 0.2 0.1 3 0.2 0.4 0.4 4 0.1 0.3 0.6 5 0.8 0.1 0.1 6 0.5 0.2 0.3 7 0.6 0.3 0.1 8 0.2 0.2 0.6 9 0.7 0.2 0.1 10 0.4 0.1 0.5

[0201] The initial cluster centers V are:

[0202]

[0203]

[0204] Iterate the cluster centers V and the membership U continuously until the termination condition is reached: the change in the objective function is less than ε or the number of iterations reaches MAX ITER .

[0205] Let m = 2, λ = 1, δ 2 = 0.5, η = 0.1, p = 1. After 20 iterations,

[0206] The objective function J decreases from the initial 8.37 to 0.58, which is less than ε. At this time, the optimized membership matrix U and the cluster centers V are as follows:

[0207] Membership matrix U:

[0208] Sample Number Category 1 Category 2 Category 3 1 0.05 0.93 0.02 2 0.08 0.89 0.03 3 0.02 0.95 0.03 4 0.01 0.97 0.02 5 0.92 0.05 0.03 6 0.14 0.83 0.03 7 0.21 0.76 0.03 8 0.01 0.02 0.97 9 0.85 0.09 0.06 10 0.02 0.95 0.03

[0209] Cluster centers V:

[0210] Category Main Steam Pressure Main Steam Temperature Generator Power Category 1 -1.00 -1.80 -1.28 Category 2 0.39 0.34 0.36 Category 3 1.50 2.00 2.00

[0211] Operating condition division:

[0212] According to the optimized membership matrix U, the category with the maximum membership degree is taken as the operating condition to which the sample belongs. It can be seen that samples 1-4, 6-7, and 10 are classified into category 2; samples 5 and 9 are classified into category 1; sample 8 is classified into category 3. These 3 categories are the operating condition classification results of this group of steady-state operating condition data, that is, the first operating condition category. Category 1 represents the low-pressure, low-temperature, and low-load operating condition, category 2 represents the medium-pressure, medium-temperature, and medium-load operating condition, and category 3 represents the high-pressure, high-temperature, and high-load operating condition. Fuzzy C-means clustering objectively depicts the characteristics of different steady-state operating conditions by optimizing the membership degree of samples and the clustering centers of categories, and can be used as an important basis for subsequent optimization control. The objective function J comprehensively considers factors such as the distance between samples and clustering centers, the fuzziness of membership degrees, and the balance of categories, making the clustering results more reasonable. It should be noted that the setting of clustering parameters needs to be repeatedly tested and adjusted according to the specific data distribution and physical background to obtain the best operating condition classification effect.

[0213] This example shows how to use the fuzzy C-means clustering algorithm to divide multiple steady-state operating condition samples into several categories, and each category represents a typical operating condition. Operating condition classification is the basis for the overall operating condition optimization of the unit. Only by accurately depicting the characteristics of different operating conditions can an optimized control strategy be designed on this basis. Step S2200 gives an effective operating condition classification method, laying a foundation for subsequent operating condition optimization control.

[0214] Step S2300, screen the first optimal operating condition category from the first operating condition category, and extract the corresponding operating parameters under the first optimal operating condition category to form the first optimal operating condition table;

[0215] Furthermore, as Figure 5 shown, step S2300 includes:

[0216] Step S2310, for each operating condition category, calculate its various performance indicators;

[0217] Step S2320, perform a weighted sum on the various performance indicators to obtain the comprehensive performance evaluation value of each operating condition category;

[0218] Step S2330, select the operating condition category with the optimal comprehensive performance evaluation value as the first optimal operating condition category;

[0219] Step S2340, extract the typical operating parameters under the first optimal operating condition category to form the first optimal operating condition table.

[0220] Specifically, step S2300 needs to establish a scientific evaluation index system to screen the optimal operating conditions from multiple operating condition categories. Since different performance indicators may have a mutually restrictive relationship, a single indicator cannot comprehensively reflect the quality of the operating conditions, and a comprehensive evaluation is required. The performance indicators of each operating condition category include load level, heat consumption, coal consumption, NOx emissions, pressure, temperature, etc.; by calculating the statistical averages of load level, heat consumption, coal consumption, NOx emissions, pressure, temperature, etc. under each operating condition category, the performance differences of different operating conditions can be initially compared. On this basis, combined with the actual management objectives and requirements of the unit, different weights are assigned to each indicator to construct a weighted comprehensive evaluation model.

[0221] The selection of weights needs to consider various factors. From an economic perspective, the weights of heat consumption and coal consumption should be relatively high to reduce the power generation cost; from a safety perspective, the weights of pressure and temperature should be appropriate to ensure that the unit does not exceed the allowable value; from an environmental protection perspective, the weight of NOx emissions should be prominent to meet the environmental protection requirements of ultra-low emissions. When there is no clear optimization objective, the same weight coefficient can be assigned to each indicator. The rationality of the weights needs to be verified through historical data analysis and expert judgment to balance the requirements of unit efficiency, safety, and environmental protection.

[0222] After weighted summation, there may be obvious differences in the comprehensive performance evaluation values of different operating condition categories. The operating condition with the highest evaluation value usually takes into account multiple aspects of the unit's performance, represents the optimal operating state, and can be used as the target for operating condition optimization. Generally speaking, the optimal operating condition corresponds to the operating state of the unit with high efficiency, low cost, stable safety, and clean environmental protection. Extract a series of typical operating parameters from the optimal operating condition, including load, main steam pressure / temperature, reheat steam temperature, furnace negative pressure, oxygen content, primary and secondary air volumes, coal consumption, coal mill combination, feed water temperature, etc., to form an optimal operating condition table, which provides an operable reference range for real-time operating condition optimization.

[0223] Exemplarily, assume that FCM clustering divides the unit operating conditions into three categories: high-load operating condition (A1), low-load operating condition (A2), and peak shaving operating condition (A3), and the performance indicators of each operating condition are as follows:

[0224] Operating condition A1: average load 95% of the rated value, power supply coal consumption 280 g / kWh, NOx emissions 30 mg / m3;

[0225] Operating condition A2: average load 60% of the rated value, power supply coal consumption 320 g / kWh, NOx emissions 80 mg / m3;

[0226] Operating condition A3: average load 80% of the rated value, power supply coal consumption 300 g / kWh, NOx emissions 100 mg / m3.

[0227] If two indicators of coal consumption and NOx emissions are considered, with weights of 0.6 and 0.4 respectively, the comprehensive performance evaluation values of the three working conditions are as follows:

[0228] Working condition A1: 280×0.6 + 30×0.4 = 180;

[0229] Working condition A2: 320×0.6 + 80×0.4 = 224;

[0230] Working condition A3: 300×0.6 + 100×0.4 = 220.

[0231] It can be seen that the comprehensive evaluation value of working condition A1 is the lowest, which is the optimal working condition. Key parameter values such as load, main steam temperature, and coal mill combination are extracted from it to form an optimal working condition operation parameter table:

[0232] Load: 950MW ± 50MW;

[0233] Main steam pressure: 24MPa ± 1MPa;

[0234] Main steam temperature: 560°C ± 10°C;

[0235] ……

[0236] This optimal working condition table clarifies the typical parameter setting ranges under the optimal operating state of the unit, which can guide the operators to optimize the working conditions. When the real-time data deviates from the setting ranges of the optimal working condition table, the relevant parameters can be adjusted in a timely manner to make the unit working condition approach the optimal direction and achieve the purpose of optimized control of the working conditions.

[0237] Step S3000, perform on-line working condition judgment and optimization for the second processed data;

[0238] Furthermore, step S3000 includes:

[0239] Step S3100, perform a steady-state judgment on the second processed data. If it is in a steady state, execute step S3200; if it is not in a steady state, keep the first optimal working condition table unchanged and execute step S3300;

[0240] Specifically, step S3100 uses the same time series steady-state determination method as step S2000 to realize the stability judgment of the unit's real-time working conditions. The timeliness requirement of on-line steady-state judgment is higher, and it is necessary to timely reflect the dynamic changes of the unit's working conditions to provide a basis for optimized control. Therefore, step S3100 selects the latest 24-hour second processed data for sliding window calculation to obtain the second steady-state index curve SI max1 (t). By setting the second steady-state threshold SI th1With the continuous time threshold T1, it can be determined whether the unit operating condition at the current moment is in a stable operation state. If the current condition is in a steady state, it indicates that the unit is operating smoothly and the operating condition optimization control can be carried out; if it is in a non-steady state, it means that the unit is in the process of operating condition conversion or fluctuation, and it is not suitable to optimize. At this time, the original first optimal operating condition table remains unchanged, and the optimization is carried out after the operating condition becomes stable. This real-time steady state judgment strategy can avoid the violent fluctuation of the unit operating condition caused by the optimization control and ensure the safe and stable operation of the unit.

[0241] Step S3200: Classify the second processed data to obtain the second operating condition category; compare the second operating condition category with the first optimal operating condition table. If the second operating condition category is better than the first optimal operating condition category, update the first optimal operating condition table with the operating parameters corresponding to the second operating condition category to obtain the second optimal operating condition table; if the second operating condition category is not better than the first optimal operating condition category, keep the first optimal operating condition table unchanged, and then execute step S3300.

[0242] Specifically, step S3200 updates the optimal operating condition knowledge base online in real time, reflecting the self-learning function of the operating condition. After the current unit operating condition is determined to be in a steady state, the characteristics of its process parameters in the parameter space can be extracted and classified using the same fuzzy C-means clustering algorithm as in step S2200 to obtain the operating condition category to which the steady state operating condition belongs. By comparing with each operating condition category in the first optimal operating condition table one by one, the optimal operating condition most similar to the current unit operating condition is found. If the current operating condition is better than the existing optimal operating condition, the operating parameters corresponding to it can be used to update the optimal operating condition table to obtain the second optimal operating condition table. The "better than" criterion here can be determined according to the unit performance indicators, such as heat consumption, coal consumption, power generation efficiency, pollutant emissions, etc. If there are multiple criteria, a comprehensive performance index can also be obtained by weighting them as a measure of the quality of the operating condition. Through this real-time comparison and learning, the best operating conditions of the unit can be continuously discovered and solidified, forming a mapping relationship between the configuration parameters and performance indicators in the knowledge base, providing decision support for real-time optimization control. On the one hand, this process makes full use of the historical optimal operating condition knowledge of the unit, and on the other hand, it can also dynamically adapt to the changing trend of the operating condition, continuously iterating the operating condition optimization model in inheritance and innovation to improve the adaptability of the unit. If the current operating condition is equivalent to or worse than the existing optimal operating condition, there is no need to update the optimal operating condition table, and the first optimal operating condition table remains unchanged as the target operating condition set for optimization control.

[0243] Step S3300: After waiting for 24 hours, store the latest 24-hour data as the new second historical operation data in the real-time database, and return to step S3100 to perform the next round of online operating condition judgment and optimization.

[0244] Specifically, step S3300 realizes the rolling update and closed-loop control of online operating condition judgment. After each decision cycle ends, the operating data of the unit for a new cycle will be added to the real-time database. The latest 24-hour data is used as the new second historical operating data to replace the data of the previous cycle and participate in the next round of steady-state judgment and operating condition benchmarking. At the same time, the second historical operating data of the previous cycle will enter the historical database and become a sample for the long-term operating condition analysis of the unit. By continuously iterating steps S3100 to S3300, the dynamic closed-loop control of operating condition optimization can be achieved, that is, according to the online operating condition judgment result, the optimal operating condition knowledge base is automatically updated, a real-time optimization control strategy is generated, the unit operating condition is adjusted, and the optimization effect is evaluated, and then enter the next optimization cycle. This "identification - decision - execution - evaluation" loop structure can continuously explore the optimization potential of the unit operation, online verify and correct each operating condition optimization strategy, continuously accumulate new operating condition knowledge, build an operating condition big data analysis platform, and provide experience support for the whole life cycle management of the unit.

[0245] Exemplarily, assume that a supercritical thermal power unit has been operating continuously and stably for 60 minutes, triggering online steady-state judgment and operating condition optimization. The operating data of the unit for these 60 minutes is as follows:

[0246]

[0247] The second steady-state index SI for these 60 minutes max1 is less than the set second steady-state threshold SI th1 , and it is determined that the current unit is in a steady-state operating condition. Cluster analysis is performed on the parameters of this steady-state operating condition, and it is found that it belongs to operating condition B1. Query the first optimal operating condition table as follows:

[0248]

[0249]

[0250] Compare the current unit operating condition with the first optimal operating condition table, and it is found that it is closest to operating condition B1, and the heat consumption of the unit is lower than that of operating condition B1. Therefore, update operating condition B1 with the current unit operating condition to obtain the second optimal operating condition table:

[0251]

[0252] When the next 24-hour cycle arrives, replace the above 60-minute data with the latest 24-hour data, start a new round of steady-state judgment and operating condition optimization, and continuously update the operating condition knowledge base.

[0253] In summary, through the interaction between the real-time database and the historical database, the method of the present invention realizes the combination of online optimization and offline analysis of the unit operating conditions. The operating condition knowledge can be learned online, accumulated offline, applied online, and updated offline, forming a data-driven closed-loop optimization control system. Compared with the traditional offline optimization and fixed-period online optimization, this method can significantly improve the optimization timeliness, shorten the knowledge iteration cycle, and enhance the dynamic adaptability of the unit operating conditions. The present invention provides a new idea for the intelligent optimized operation of thermal power units, which is of great significance for energy conservation, emission reduction, and efficiency improvement.

[0254] Embodiment 2

[0255] Based on Embodiment 1, this embodiment provides an operating condition optimization system for a thermal power unit, as Figure 6 shown, including:

[0256] Data acquisition module: used to obtain the first historical operation data and the second historical operation data of the unit;

[0257] Data processing module: used to perform anomaly processing on the obtained first historical operation data and the second historical operation data to obtain the first processed data and the second processed data;

[0258] Steady-state screening module: used to perform steady-state screening on the first processed data to obtain the first steady-state data;

[0259] Operating condition division module: used to perform operating condition division on the first steady-state data by using the fuzzy C-means clustering algorithm to obtain the first operating condition category;

[0260] Optimal operating condition screening module: used to screen out the first optimal operating condition category from the first operating condition category and extract the corresponding operating parameters under the first optimal operating condition category to form the first optimal operating condition table;

[0261] Online operating condition optimization module: used to perform online operating condition judgment and optimization on the second processed data and update the first optimal operating condition table.

[0262] Embodiment 3

[0263] This embodiment discloses an electronic device 500, as Figure 7 shown. The electronic device 500 may include one or more processors and one or more memories. Among them, computer-readable code is stored in the memory. When the computer-readable code is run by one or more processors, it can execute an operating condition optimization method for a thermal power unit operation as described above.

[0264] The method or system according to the embodiments of the present application can also be implemented by means of Figure 7 the architecture of the electronic device shown. As Figure 7As shown, the electronic device 500 may include a bus 501, one or more CPUs 502, a read-only memory (ROM) 503, a random access memory (RAM) 504, a communication port 505 connected to a network, an input / output component 506, a hard disk 507, and so on. The storage device in the electronic device 500, such as the ROM 503 or the hard disk 507, may store a method for optimizing operating conditions of a thermal power unit provided in this application. A method for optimizing operating conditions of a thermal power unit may, for example, include: obtaining first historical operation data and second historical operation data of the unit, performing anomaly processing on the obtained first historical operation data and second historical operation data to obtain first processed data and second processed data; performing steady-state screening on the first processed data to obtain first steady-state data; using the fuzzy C-means clustering algorithm to perform operating condition classification on the first steady-state data to obtain a first operating condition category, screening out a first optimal operating condition category from the first operating condition category, and extracting the operating parameters corresponding to the first optimal operating condition category to form a first optimal operating condition table; performing online operating condition judgment and optimization on the second processed data to update the first optimal operating condition table. Further, the electronic device 500 may further include a user interface 508. Of course, Figure 7 The architecture shown is only exemplary. When implementing different devices, one or more components in the electronic device shown may be omitted according to actual needs. Figure 7 shown may be omitted according to actual needs.

[0265] Embodiment 4

[0266] This embodiment discloses a computer-readable storage medium 600, as Figure 8 shown. Computer-readable instructions are stored on the computer-readable storage medium 600. When the computer-readable instructions are run by a processor, a method for optimizing operating conditions of a thermal power unit according to an embodiment of this application described with reference to the above drawings may be executed. The storage medium 600 includes but is not limited to, for example, volatile memory and / or non-volatile memory. Volatile memory may, for example, include random access memory (RAM) and cache memory, etc. Non-volatile memory may, for example, include read-only memory (ROM), hard disk, flash memory, etc.

[0267] In addition, according to an embodiment of the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the present application provides a non-transitory machine-readable storage medium storing machine-readable instructions that can be run by a processor to execute instructions corresponding to the method steps provided by the present application, such as: obtaining first historical operation data and second historical operation data of the unit, performing exception handling on the obtained first historical operation data and second historical operation data to obtain first processed data and second processed data; performing steady-state screening on the first processed data to obtain first steady-state data; using the fuzzy C-means clustering algorithm to perform operating condition classification on the first steady-state data to obtain a first operating condition category, screening out a first optimal operating condition category from the first operating condition category, and extracting operating parameters corresponding to the first optimal operating condition category to form a first optimal operating condition table; performing online operating condition judgment and optimization on the second processed data to update the first optimal operating condition table. When the computer program is executed by a central processing unit (CPU), the above functions defined in the method of the present application are executed.

[0268] The methods, devices, and equipment of the present application can be implemented in many ways. For example, the methods, devices, and equipment of the present application can be implemented through software, hardware, firmware, or any combination of software, hardware, and firmware. The above order of steps for the method is only for illustration, and the steps of the method of the present application are not limited to the specific order described above, unless otherwise specifically stated. In addition, in some embodiments, the present application can also be implemented as a program recorded in a recording medium, and these programs include machine-readable instructions for implementing the method according to the present application. Therefore, the present application also covers a recording medium storing a program for executing the method according to the present application.

[0269] In addition, parts of the above technical solutions provided in the embodiments of the present application that are consistent with the implementation principles of the corresponding technical solutions in the prior art are not described in detail to avoid excessive elaboration.

[0270] As described above in the specific embodiments, the purpose, technical solutions, and beneficial effects of the present invention have been further described in detail. It should be understood that the above are only specific embodiments of the present invention and are not used to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for optimizing the operating conditions of a thermal power unit, characterized in that, The method includes: Obtaining the first historical operation data and the second historical operation data of the unit, performing anomaly processing on the obtained first historical operation data and second historical operation data to obtain the first processed data and the second processed data; Performing steady-state screening on the first processed data to obtain the first steady-state data; using the fuzzy C-means clustering algorithm to perform working condition classification on the first steady-state data to obtain the first working condition category, screening out the first optimal working condition category from the first working condition category, and extracting the corresponding operation parameters under the first optimal working condition category to form the first optimal working condition table; Performing online working condition judgment and optimization on the second processed data, and updating the first optimal working condition table; The performing steady-state screening on the first processed data to obtain the first steady-state data includes: Extracting the time series data of the steady-state parameters from the first processed data, dividing the time series data of each steady-state parameter into sliding time windows with a length of L, and the windows slide continuously at a step size of 1 sampling point; Calculating the steady-state index SI for the data within each time window of each steady-state parameter; Calculate SI for multiple steady-state parameters within each time window respectively, and take the maximum value SI of each SI max as the comprehensive steady-state index for this time window; Take the average of the timestamps of all sampling points within the sliding time window as the comprehensive steady-state index SI max For the corresponding moment t, obtain the steady-state index curve SI of the time series max (t); Set the first steady-state threshold SI th , and find SI max in all time intervals T th less than SI s , and if T s is greater than the interval threshold T, then determine that this interval is a steady-state interval; extract the first processed data within the steady-state interval as the first steady-state data; The method for calculating the steady-state index SI includes: where x i is the value of the i-th sample data point, μ is the average value of all sample data points within the time window, σ is the standard deviation of all sample data points within the time window, w i is the weight of the i-th sample data point, κ is the adjustment coefficient, ν is the change rate of data points within the time window, and N is the number of sample data points within the time window.

2. The method for optimizing the operating conditions of a thermal power unit according to claim 1, wherein, The first historical operation data is the historical operation data of the unit in the most recent 12 months obtained from the real-time database, and the sampling interval is 1 minute; the second historical operation data is the operation data within the previous 24 hours obtained from the real-time database, and the sampling interval is 1 second.

3. The optimization method for operating conditions of a thermal power unit according to claim 1, characterized in that, The performing anomaly processing on the obtained first historical operation data and second historical operation data includes: Performing missing value detection item by item on the first historical operation data and the second historical operation data. If the data is missing, it is marked as missing data. If the data is not missing, it is marked as non-missing data; for the missing data, it is replaced with the average value of the 5 adjacent data before and after the adjacent moment; if there are 2 or more consecutive missing data in the adjacent data, it is marked as unavailable data; Performing physical threshold detection item by item on the non-missing data, setting the physical threshold range. If it exceeds the physical threshold range, it is marked as the first abnormal data; Performing statistical threshold detection item by item on the data that does not exceed the physical threshold range. Taking 3 times the standard deviation as the statistical threshold, if the deviation from the data average value exceeds 3 times the standard deviation, it is marked as the second abnormal data; Performing noise detection item by item on the data that has not been detected as abnormal, calculating the peak-to-peak value of the data. If the peak-to-peak value exceeds the set peak-to-peak threshold and the short-term variance of the data is higher than 3 times the long-term variance, it is marked as the third abnormal data; Removing the unavailable data, the first abnormal data, the second abnormal data and the third abnormal data to obtain the processed first processed data and second processed data.

4. A method for optimizing the operating conditions of a thermal power unit according to claim 1, characterized in that, The performing working condition classification on the first steady-state data to obtain the first working condition category includes: Performing data standardization processing on the first steady-state data to obtain the standardized steady-state data; Initialize the parameters of the fuzzy C-means clustering algorithm, including the number of clusters C, the membership matrix U, the cluster centers V, and the maximum number of iterations MAX ITER , and the termination threshold ε of the objective function; Based on the standardized steady-state data, iteratively optimizing the membership matrix U and the clustering center V; According to the final membership matrix U, dividing each data point into the category with the largest membership to form C working condition clusters, that is, the first working condition category; The iteratively optimizing the membership matrix U and the clustering center V includes: Update the cluster center V for each category based on the current membership matrix U k , where k is the index of the number of cluster categories, and V k represents the cluster center of the k-th category; Update the membership degrees U of data points to each category based on the current cluster center V kj , U kj represents the membership degree of the j-th sample belonging to the k-th category; Calculate the objective function J and determine whether J is less than ε or the number of iterations has reached MAX ITER , if so, generate the final membership matrix U; if not, continue the iteration.

5. A method for optimizing operating conditions of a thermal power unit according to claim 1, characterized in that, The screening out the first optimal working condition category from the first working condition category includes: For each operating condition category, calculate its various performance indicators; Perform a weighted sum of the various performance indicators to obtain the comprehensive performance evaluation value for each operating condition category; Select the operating condition category with the optimal comprehensive performance evaluation value as the first optimal operating condition category.

6. The operating condition optimization method for a thermal power unit according to claim 1, characterized in that, The online operating condition judgment and optimization for the second processed data include: Perform a steady-state judgment on the second processed data. If it is in a steady state, perform an operating condition division on the second processed data to obtain the second operating condition category; compare the second operating condition category with the first optimal operating condition table. If the second operating condition category is better than the first optimal operating condition category, update the first optimal operating condition table with the operating parameters corresponding to the second operating condition category; if the second operating condition category is not better than the first optimal operating condition category, keep the first optimal operating condition table unchanged. After waiting for 24 hours, store the latest 24-hour data as the new second historical operating data in the real-time database and perform the next round of online operating condition judgment and optimization; If it is not in a steady state, keep the first optimal operating condition table unchanged. After waiting for 24 hours, store the latest 24-hour data as the new second historical operating data in the real-time database and perform the next round of online operating condition judgment and optimization.

7. A working condition optimization system for a thermal power unit operation, which is used to implement the working condition optimization method for a thermal power unit operation described in any one of claims 1-6, characterized in that, The system includes: A data acquisition module: used to obtain the first historical operating data and the second historical operating data of the unit; A data processing module: used to perform anomaly processing on the obtained first historical operating data and the second historical operating data to obtain the first processed data and the second processed data; A steady-state screening module: used to perform steady-state screening on the first processed data to obtain the first steady-state data; An operating condition division module: used to perform operating condition division on the first steady-state data using the fuzzy C-means clustering algorithm to obtain the first operating condition category; An optimal operating condition screening module: used to screen out the first optimal operating condition category from the first operating condition category and extract the corresponding operating parameters under the first optimal operating condition category to form the first optimal operating condition table; An online operating condition optimization module: used to perform online operating condition judgment and optimization for the second processed data and update the first optimal operating condition table.

8. An electronic device, comprising a memory, a central processing unit, and a computer program stored on the memory and executable on the processor, characterized in that When the processor executes the computer program, it implements the steps in the method for optimizing the operating conditions of a thermal power unit according to any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, A computer program is stored on the computer-readable storage medium, and when the computer program is executed, it implements the steps in the method for optimizing the operating conditions of a thermal power unit according to any one of claims 1-6.

Images